iiinova — where your digital self keeps evolving

I see you, Ashley Sanchez!!!!!!!!! Hell yeah! @nccourage.com

011

One thing I still, to this day, do not quite understand... is what kind of work everyone else is doing who really think they *need* Opus/Astra level of AI? Hell, even Luna/Sonnet level. I'm just over here happy as a puppy with a new toy with GLM 5.3 Flash and DSV4.1 Flash, still.

000

Finally got around to getting Excalidraw set up on the server. There's something quietly endearing about prompting an agent in Hermes with very little ("a drawing inspired by the movie Interstellar") and just seeing what happens.

The day after Muse shared users text messages? I don't get the people who would sign up for this.

Man, Mr. Belding died. RIP.

110

As someone who has been unashamedly using tools like OpenClaw and Hermes since *January,* I have a lot of opinions. I am a C- at best when it comes to my own op-sec, but *somehow* things like this have not happened to me. Not once. I have opinions on why, not sure people will like them though.

000

Small thing, but I am annoyed by nobody doing proper versioning anymore. What happened to 5.1-5.4?!

000

Absolutely nobody (everyone) saw this coming.

000

Over the weekend, I split up some inference over my potato M1 Mac Mini 16 GB and 9070 XT powered PC to make Sub/Wave entirely self hosted. Gemma 4 E4B is surprisingly great at writing the DJ scripts, and Breeze TTS 2 is mind-blowing-ly good at TTS. Sure it could be faster via API...

3141

I've been using Jev now for a week or two (however long its been since it came out). I've spent all of...

Joshua White @jrw14.whnc.me · 26/09/2026

Life really does never get any easier, does it?

100

Voice upgrade. Breeze TTS 2 is very good. Set up a bridge between Sub/Wave (server) and Breeze (running on a pc previously set up for gaming) with hard-coded voice settings. It takes 3x as long as Kokoro running in the same container, but since it's got 3-5 minutes (song) it isn't noticeable.

051

Nicholas Schnittker @nick-schnittker.bsky.social · 25/09/2026

I've been working on setting up a new TTS engine (Breeze 2 is incredible, but still optimizing it). In the meantime, this is what the DJ transitions sound like. Honestly, pretty freaking good for Kokoro's 82m STT engine (af_heart). Completely automated.

061

Reposted by Joshua White

A must-read for Courage Country. An excellent look behind the scenes at some of what has led to Sanchez's huge 2026 and the influence Mak Lind has had on unlocking the best version of her on the field.

1122

I keep talking about this, because honestly it's made me re-think my homelab a bit. For the last year I have been almost exclusively focused on data sovereignty. Yes it's rewarding and important, but it's not always "fun." And lately, with work, decision-fatigue is a thing for me. Sub/Wave = joy.

250

In the age of agentic coding, one thing I've taken for granted most of my adult-life has become something I am deeply grateful for... The ability to speed-read. 😂

070

Joshua White @jrw14.whnc.me · 24/09/2026

Spent a little more time on BubbleRadio tonight. Added support for themes, album art, dj avatars, dynamic re-sizing, and cleaned up several bugs. Including reducing resource usage pretty significantly. TUI's are magic. (Sub/wave, Go, Charm's Bubbletea)

020

These last two weeks at work have felt like an entire year.

static.klipy.com

Imtired Boss

ALT: Imtired Boss

000

Joshua White @jrw14.whnc.me · 24/09/2026

Had some fun with @charm.land's Go goodies tonight and made a custom TUI audio player for SUB/WAVE complete with real-time spectrum analyzer. Linux distro: Endeavour OS Shell: Noctalia WM: Niri Charm goodies: bubbletea, lipgloss, bubbles, harmonica

252

Joshua White @jrw14.whnc.me · 23/09/2026

A testament to the level of disinformation and political discourse surrounding AI in the US.

161

Yes, this is 100% more exciting to me than a new Claude model.

0130

GitHub - mizorewww/laya-mlx: Native MLX runtime for Laya typed decision models — 7–14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API.

Gosh, I really love the opensource community. This morning I was thinking to myself, if my potato M1 Mac Mini was worth keeping on for just an embedding model and ling-tiny-3.0 LLM (for very light background tasks). Then I found github.com/mizorewww/la.... Faster and soveriegn Jev? Yes, please.

github.com

Native MLX runtime for Laya typed decision models — 7–14 ms short decisions on M3 Max. No text generation, PyTorch, or cloud API. - mizorewww/laya-mlx

2190

It's a small thing many might not see the value in, but Sub/wave, the FOSS, self-hosted, AI-powered internet radio station is still impressing me days later and is amongst my very favorite things I've put in my homelab.

000

Joshua White @jrw14.whnc.me · 21/09/2026

<OpenAI, Anthropic, xAI, Google>: We should slow down and self-regulate. <Also OpenAI, Anthropic, xAI, Google>: All releasing new models this week. 😵‍💫 Makes me think the concern re: safety may not really be about safety after all, at least not concerning their own models.

380

Xiaomi has arrived to eat every other model's lunch and dinner. Holy crap.

0110

Joshua White @jrw14.whnc.me · 21/09/2026

Well, if nothing else, the Bonsai local LLM experiment at least got me off my rear to experiment hosting models locally. I never thought to do image generation local, but this machine is kind of perfect for it. Get images in about 10 seconds, which is a-okay by me.

1140

Joshua White @jrw14.whnc.me · 21/09/2026

Eating some crow tonight. Remember that coding harness showdown I ran about a week ago? I tried it again tonight. Just one harness (Pi), but switched out the model from GLM 5.3 Flash to Bonsai 2 27b. And, it wasn't pretty.

170

That's that then. Set up my gaming pc to be always available (via wake on lan), and added it as a full-time (as needed) host in the homelab. I run inference at home now. Wild. I figure I'll use it for intentional sessions, not automated background-work, that way I can keep it off most of the time.

140

THAT SAID... with kv caching and ttft under 500ms, "real feel" absolutely is around 45-50 tps. It does not feel "slow" by any means.

130

Increased Bonzai 2 27b's tokens per second by 30% by enabling KV caching. Haven't been able to hit 50 tps yet... but there's still some juice left to squeeze here. Having to use a fork of Prism ML's llama.cpp (AMD) isn't helping either, as there are no speculative decoders available.

0100

Joshua White @jrw14.whnc.me · 19/09/2026

Yep. They played really well though. In other news, EMP is electric and is quickly becoming a fan favorite.

002

Okay. Uh. Yeah. Bonsai 2 27b is making my life difficult. It's good. Too good to ignore. It runs _fine_ on my 9070XT. I did zero optimization of prompt-caching, and was still hitting a sustained 30 tp/s with 128k context on a 20 minute small coding job. I think I can get to 50 w/ optimizing.

3190

Joshua White @jrw14.whnc.me · 19/09/2026

My biggest question going into setting up Sub/Wave today was what token consumption would look like. I wanted a "best" baseline, so turned everything on; agentic song picking, aggressive dj segments, medium length inserts, etc... And after a LOT of hours left on, I'm kinda shocked < 1 million tks.

000

Joshua White @jrw14.whnc.me · 18/09/2026

Uh. Say what now?! This might be the thing that gets me to add my gaming pc to the homelab as AI-inference instead of gaming, ha. It's not much, 32gb RAM with AMD's 9800X3D and 9070XT GPU (16gb), but... if I'm reading all this right, it should be enough to still expect 50ish tp/s.

450

Joshua White @jrw14.whnc.me · 18/09/2026

Accidentally wading into the mess that is Omarchy turned out to be pretty solid for me as it ultimately introduced me to Niri and Noctalia. Once I realized who DHH was, I ran from Omarchy as fast as I could, and turns out Niri & Noctalia with EndeavourOS feel MUCH more at home to my brain.

090

Joshua White @jrw14.whnc.me · 18/09/2026

I do a lot of cool things with my home lab that mostly revolve around self-hosting and data sovereignty. But every once in a while, you just need a new toy to play with. Sub/wave might be the coolest thing I've put on my server yet. github.com/perminder-kl...

041

Joshua White @jrw14.whnc.me · 17/09/2026

I don't know about y'all, but I've been busy wiring Jev into every single decision-making workflow I've got 😂

270

Joshua White @jrw14.whnc.me · 16/09/2026

Union Alpha smells like a smaller lab to me. Explicitly not being trained on during their Openrouter stealth run... That has small shop, open weight provider written all over it. The smaller context size adds to this. Perhaps a GLM 5 Air? Poolside? Arcee? Qwen? Dark horse: new GPT-OSS?

040

Joshua White @jrw14.whnc.me · 14/09/2026

Time to start streaming Macklemore again.